A New Approach for Performance Evaluation of AOGCM Models in Simulating Runoff

Background and objectives: phenomena of climate change affects on various sectors that water resources are those most important. Lane et al (1999) studied that countries located in low-latitude, have the most the negative consequences of this phenomenon (9). A prerequisite for evaluating the regional effects of climate change is to produce climatic scenarios in the future period’s by AOGCM models. Different researchers depending on their needs, the outputs of one or more of these models use. Yu et al (2002) examined effects of climate change on water resources in southern Taiwan with the use of models (12). Wilby and Harris (2006) studied the effect of climate change on low flows of the river in the United Kingdom with considering the uncertainties of AOGCM models, along with other sources of uncertainty, (11). The aim of the present study is to evaluate the performance of climate models from two perspectives of hydrological and water resources. This means that in addition to determination of situation of the system probability distribution in each month (perspective of hydrologists), the time sequence of the system situation to be evaluated under climate change conditions.
Materials and Methods: In the present study, a new approach was introduced for the performance investigation of AOGCM models, so that reliable model(s) could be found with saving time and obtaining satisfactory results. With applying 7 AOGCMs, temperature and rainfall variables in base period (1971-2000) for Aidoghmoush basin located in East Azerbaijan were estimated and with introducing variables to hydrological model IHACRES, monthly runoff was simulated. To investigate efficiency of each climate model, the mean observed runoff method was used. Next, a hybrid model was also suggested, so as to assign the higher values to each model in each month. By fitting statistical distributions on runoff and using goodness-of-fit tests, an appropriate distribution was chosen and relevant statistical parameters extracted and compared with observed runoff.
Results: Results show that the hybrid and HadCM3 models with respective (r = 96%, RMSE = 2.09 m3/s, MAE = 1.51 m3/s, NSE = 0.89) for the hybrid model and (r = 97%, RMSE = 2.32 m3/s, MAE = 1.58 m3/s, NSE = 0.87) for HadCM3 can best simulate the runoff. Next, transition probability matrix was assessed. By comparing the results of probability distributions and the transition probability matrix for runoff resulting from AOGCMs with observed runoff showed that performance of hybrid and HadCM3 models with respective correlation coefficients of 89% (RMSE = 0.1 m3/s, MAE = 0.02 m3/s, NSE = 0.78) and 87% (RMSE = 0.12 m3/s, MAE = 0.02 m3/s, NSE = 0.77) can be reliable.
Conclusion: Models have Good ability to simulate climatic variables and consequently runoff. In case of using only one AOGCM model, since there is no significant difference between performance criteria of hybrid and HadCM3 models, applying the HadCM3 is recommended. The results of comparisons of statistical and flow transition probability are quite similar.